February 2025 · 9 min read

How to Spot Fake News Using Forensic Tools

The spread of fake news has accelerated dramatically in the age of AI-generated media. Images, videos and text that would have taken days to fabricate convincingly can now be produced in seconds using publicly available AI tools. Developing a robust fact-checking workflow is essential for anyone who publishes or shares information online.

Step 1 — Reverse image search

Before accepting any image at face value, perform a reverse image search using Google Images, TinEye or Yandex Images. Upload the image or paste its URL and search for other instances of it online. If the image appears in other contexts with different captions or dates, this is a strong indicator that it is being misused or manipulated.

Step 2 — Check image metadata

Examine the EXIF metadata embedded in the image file. This can reveal the camera model, the date and time the photograph was taken, and in some cases the GPS location. Discrepancies between the claimed date or location and the metadata are a significant red flag. Completely absent metadata in what is claimed to be a news photograph is also suspicious, since most modern cameras embed metadata automatically.

Step 3 — AI forensic analysis

Use a dedicated forensic analysis tool to determine whether an image or video has been AI-generated or manipulated. Chicken AI examines 47 detection dimensions including GAN fingerprints, pixel coherence, noise patterns and compression artefacts. The tool is free and produces results in under 3 seconds with a detailed breakdown of forensic signals.

Step 4 — Cross-reference with primary sources

No forensic tool should be the only line of verification. Always attempt to trace a claim back to its primary source — the original event, document or recording. Check whether reputable news organisations have reported on the same event. Look for corroborating evidence from multiple independent sources before drawing conclusions or sharing information.

Step 5 — Check the text for AI generation

AI-generated text articles can be identified by several characteristics. They tend to have very uniform sentence length, lack specific names and details, use hedging language excessively, and fail to cite verifiable sources. Chicken AI's text analysis domain can assess written content for AI generation indicators including perplexity and burstiness scores — measures of how predictable and variable the text is compared to human writing.